o %ݫiw@@sdZddlZddlmZddlmmZddlZddl m Z dZ Gdddej Z Gdddej ZGd d d ejjjZGd d d ej ZGd ddejj ZGdddejjjZGdddej ZddZGdddej ZGdddej ZdS)z6 Implementation of a popular speech separation model. N)overlap_and_addg:0yE>c(eZdZdZfddZddZZS)EncoderaThis class learns the adaptive frontend for the ConvTasnet model. Arguments --------- L : int The filter kernel size. Needs to be an odd number. N : int Number of dimensions at the output of the adaptive front end. Example ------- >>> inp = torch.rand(10, 100) >>> encoder = Encoder(11, 20) >>> h = encoder(inp) >>> h.shape torch.Size([10, 20, 20]) cs,ttjjjd|||ddd|_dS)NF) in_channels out_channels kernel_sizestridebias)super__init__sbnnetCNNConv1dconv1d_UselfLN __class__X/home/ubuntu/.local/lib/python3.10/site-packages/speechbrain/lobes/models/conv_tasnet.pyr !s  zEncoder.__init__cCs$t|d}||}t|}|S)a Arguments --------- mixture : torch.Tensor Tensor shape is [M, T]. M is batch size. T is #samples Returns ------- mixture_w : torch.Tensor Tensor shape is [M, K, N], where K = (T-L)/(L/2)+1 = 2T/L-1 )torch unsqueezerFrelu)rmixtureconv_out mixture_wrrrforward-s  zEncoder.forward__name__ __module__ __qualname____doc__r r# __classcell__rrrrrs  rcr)DecoderaFThis class implements the decoder for the ConvTasnet. The separated source embeddings are fed to the decoder to reconstruct the estimated sources in the time domain. Arguments --------- L : int Number of bases to use when reconstructing. N : int Input size Example ------- >>> L, C, N = 8, 2, 8 >>> mixture_w = torch.randn(10, 100, N) >>> est_mask = torch.randn(10, 100, C, N) >>> Decoder = Decoder(L, N) >>> mixture_hat = Decoder(mixture_w, est_mask) >>> mixture_hat.shape torch.Size([10, 404, 2]) cs*t||_tjjj||dd|_dS)NF) input_size n_neuronsr )r r rrrlinearLinear basis_signalsrrrrr Ws  zDecoder.__init__cCsZt|ddd|dd|}|dddd}||}t||jd}|dddS)a) Arguments --------- mixture_w : torch.Tensor Tensor shape is [M, K, N]. est_mask : torch.Tensor Tensor shape is [M, K, C, N]. Returns ------- est_source : torch.Tensor Tensor shape is [M, T, C]. rrr)rrrepeatsizepermuter/rr)rr"est_masksource_w est_sourcerrrr#bs zDecoder.forwardr$rrrrr*?s  r*cs eZdZdZfddZZS)TemporalBlocksSequentialam A wrapper for the temporal-block layer to replicate it Arguments --------- input_shape : tuple Expected shape of the input. H : int The number of intermediate channels. P : int The kernel size in the convolutions. R : int The number of times to replicate the multilayer Temporal Blocks. X : int The number of layers of Temporal Blocks with different dilations. norm_type : str The type of normalization, in ['gLN', 'cLN']. causal : bool To use causal or non-causal convolutions, in [True, False]. Example ------- >>> x = torch.randn(14, 100, 10) >>> H, P, R, X = 10, 5, 2, 3 >>> TemporalBlocks = TemporalBlocksSequential( ... x.shape, H, P, R, X, 'gLN', False ... ) >>> y = TemporalBlocks(x) >>> y.shape torch.Size([14, 100, 10]) c s^tj|dt|D]!}t|D]} d| } |jt||dd| ||d|d| d qq dS)N input_shaperrsametemporalblock__rr r paddingdilation norm_typecausal layer_name)r r rangeappend TemporalBlock) rr9HPRXr@rArxr?rrrr s"  z!TemporalBlocksSequential.__init__r%r&r'r(r r)rrrrr7}s r7cs0eZdZdZ   d fdd ZddZZS) MaskNetu{ Arguments --------- N : int Number of filters in autoencoder. B : int Number of channels in bottleneck 1 × 1-conv block. H : int Number of channels in convolutional blocks. P : int Kernel size in convolutional blocks. X : int Number of convolutional blocks in each repeat. R : int Number of repeats. C : int Number of speakers. norm_type : str One of BN, gLN, cLN. causal : bool Causal or non-causal. mask_nonlinear : str Use which non-linear function to generate mask, in ['softmax', 'relu']. Example ------- >>> N, B, H, P, X, R, C = 11, 12, 2, 5, 3, 1, 2 >>> MaskNet = MaskNet(N, B, H, P, X, R, C) >>> mixture_w = torch.randn(10, 11, 100) >>> est_mask = MaskNet(mixture_w) >>> est_mask.shape torch.Size([2, 10, 11, 100]) gLNFrc sxt||_| |_t||_tjjj ||ddd|_ dd|f} t | |||||| |_ tjjj |||ddd|_ dS)NrF)rrr r )r r Cmask_nonlinearChannelwiseLayerNorm layer_normrrrrbottleneck_conv1x1r7temporal_conv_net mask_conv1x1) rrBrFrGrIrHrOr@rArPin_shaperrrr s"     zMaskNet.__init__cCs|ddd}|\}}}||}||}||}||}||||j|}|dddd}|j dkrCt j |dd}|S|j dkrOt |}|St d) aKeep this API same with TasNet. Arguments --------- mixture_w : torch.Tensor Tensor shape is [M, K, N], M is batch size. Returns ------- est_mask : torch.Tensor Tensor shape is [M, K, C, N]. rrrr0softmax)dimrz$Unsupported mask non-linear function)r3r2rRrSrTrU contiguousreshaperOrPrrXr ValueError)rr"MKryscorer4rrrr#s"        zMaskNet.forward)rNFrr$rrrrrMs+*rMcs.eZdZdZ  dfdd ZddZZS) rEaThe conv1d compound layers used in Masknet. Arguments --------- input_shape : tuple The expected shape of the input. out_channels : int The number of intermediate channels. kernel_size : int The kernel size in the convolutions. stride : int Convolution stride in convolutional layers. padding : str The type of padding in the convolutional layers, (same, valid, causal). If "valid", no padding is performed. dilation : int Amount of dilation in convolutional layers. norm_type : str The type of normalization, in ['gLN', 'cLN']. causal : bool To use causal or non-causal convolutions, in [True, False]. Example ------- >>> x = torch.randn(14, 100, 10) >>> TemporalBlock = TemporalBlock(x.shape, 10, 11, 1, 'same', 1) >>> y = TemporalBlock(x) >>> y.shape torch.Size([14, 100, 10]) rNFc st|\} } } tjjj|d|_|jjtjjj |dddd|jjt dd|jjt ||dd|jjt | ||||||d d dS) Nr8rFconvrr r rBactrBnormDSconvr=)r r rr containers SequentiallayersrDrrnnPReLU choose_normDepthwiseSeparableConv) rr9rr r r>r?r@rAr]r^rVrrrr Bs2   zTemporalBlock.__init__cCs|}||}||S)z Arguments --------- x : torch.Tensor Tensor shape is [M, K, B]. Returns ------- x : torch.Tensor Tensor shape is [M, K, B]. )ri)rrKresidualrrrr#ls zTemporalBlock.forwardrNFr$rrrrrE"s '*rEcs&eZdZdZ  dfdd ZZS)rmaBuilding block for the Temporal Blocks of Masknet in ConvTasNet. Arguments --------- input_shape : tuple Expected shape of the input. out_channels : int Number of output channels. kernel_size : int The kernel size in the convolutions. stride : int Convolution stride in convolutional layers. padding : str The type of padding in the convolutional layers, (same, valid, causal). If "valid", no padding is performed. dilation : int Amount of dilation in convolutional layers. norm_type : str The type of normalization, in ['gLN', 'cLN']. causal : bool To use causal or non-causal convolutions, in [True, False]. Example ------- >>> x = torch.randn(14, 100, 10) >>> DSconv = DepthwiseSeparableConv(x.shape, 10, 11, 1, 'same', 1) >>> y = DSconv(x) >>> y.shape torch.Size([14, 100, 10]) rNFc  stj|d|\} } } |r||d} d}d} nd} |jtjjj| ||||| dd| d |r8|jt| d d |jt d d |jt || d d |jtjjj|ddd d dS)Nr8rrAr:rFconv_0) rr r r>r?groupsr rBdefault_paddingchomprdrcconv_1rb) r r rDrrrrChomp1drjrkrl)rr9rr r r>r?r@rA batchsizetimer paddingvalrrrrrr s<    zDepthwiseSeparableConv.__init__rorLrrrrrm}s (rmcr)ruaThis class cuts out a portion of the signal from the end. It is written as a class to be able to incorporate it inside a sequential wrapper. Arguments --------- chomp_size : int The size of the portion to discard (in samples). Example ------- >>> x = torch.randn(10, 110, 5) >>> chomp = Chomp1d(10) >>> x_chomped = chomp(x) >>> x_chomped.shape torch.Size([10, 100, 5]) cst||_dS)N)r r chomp_size)rryrrrr s  zChomp1d.__init__cCs"|ddd|j ddfS)z Arguments --------- x : torch.Tensor Tensor shape is [M, Kpad, H]. Returns ------- x : torch.Tensor Tensor shape is [M, K, H]. N)ryrZ)rrKrrrr#s" zChomp1d.forwardr$rrrrrus rucCs*|dkrt|S|dkrt|St|S)aYThis function returns the chosen normalization type. Arguments --------- norm_type : str One of ['gLN', 'cLN', 'batchnorm']. channel_size : int Number of channels. Returns ------- Constructed layer of the chosen type Example ------- >>> choose_norm('gLN', 10) GlobalLayerNorm() rNcLN)GlobalLayerNormrQrj BatchNorm1d)r@ channel_sizerrrrls  rlc0eZdZdZfddZddZddZZS)rQaeChannel-wise Layer Normalization (cLN). Arguments --------- channel_size : int Number of channels in the normalization dimension (the third dimension). Example ------- >>> x = torch.randn(2, 3, 3) >>> norm_func = ChannelwiseLayerNorm(3) >>> x_normalized = norm_func(x) >>> x.shape torch.Size([2, 3, 3]) cBtttdd||_ttdd||_|dSNr r r rj ParameterrTensorgammabetareset_parametersrr}rrrr &  zChannelwiseLayerNorm.__init__cC|jjd|jjdSzResets the parameters.rNrdatafill_rzero_rrrrr,z%ChannelwiseLayerNorm.reset_parameterscCsJtj|ddd}tj|dddd}|j||t|td|j}|S)z Args: y: [M, K, N], M is batch size, N is channel size, K is length Returns: cLN_y: [M, K, N] rTrYkeepdimF)rYrunbiased?)rmeanvarrpowEPSr)rr_rrcLN_yrrrr#1s$zChannelwiseLayerNorm.forwardr%r&r'r(r rr#r)rrrrrQ  rQcr~)r{a<Global Layer Normalization (gLN). Arguments --------- channel_size : int Number of channels in the third dimension. Example ------- >>> x = torch.randn(2, 3, 3) >>> norm_func = GlobalLayerNorm(3) >>> x_normalized = norm_func(x) >>> x.shape torch.Size([2, 3, 3]) crrrrrrrr OrzGlobalLayerNorm.__init__cCrrrrrrrrUrz GlobalLayerNorm.reset_parameterscCsd|jdddjddd}t||djdddjddd}|j||t|td|j}|S)a Arguments --------- y : torch.Tensor Tensor shape [M, K, N]. M is batch size, N is channel size, and K is length. Returns ------- gLN_y : torch.Tensor Tensor shape [M, K. N] rTrrr)rrrrrr)rr_rrgLN_yrrrr#Zs $zGlobalLayerNorm.forwardrrrrrr{>rr{)r(rtorch.nnrjtorch.nn.functional functionalr speechbrainr(speechbrain.processing.signal_processingrrModulerr*rrgrhr7rMrErmrurlrQr{rrrrs"  1>3r[U')